Second LORA, had to work with much less images available than the first one, so issues were bound to happen >.<

This LORA is designed to try to recreate the semblance of Kazuko Kawakami, aka the little sister of Momoyo Kawakami (and personal "doggo" of the Kazama Family) that you can date in the Majikoi game series, critically acclaimed series of ?????? Visual Novels.

Trigger words are simple, but be careful with her hair color because the AI struggles to get the exact tone she shows in the game; use my images to understand the various combinations. On another note, this LORA suffers from two issues due to the quality of the images not being exactly optimal (she has very few compared to her big sister, and most needed an upscale):

1. The model tends, sometimes, to put censorship in the "lower regions": the solution is usually putting negative words like "censorship, censored, censored ?????, censored penis".

2. The model tends to be erratic with hands. I suggest using ADetailer to fix them (usually 0.6 strength of INpaint is more than enough), but be prepared, sometimes, to have ADetailer ??????? hands that would be perfectly fine; in that case, I suggest regenerating the image with the same seeds but turning ADetailer off.

If I can find better images for a dataset, I'll try to fix all these issues with a future release.

It works best at 0.9 weight. Trained with the PonyDiffusion model on KohyaSS, I found it produces the best images using 7th Anime.